Fusion of Remotely Sensed Displacement Measurements: Current status and challenges

Yajing Yan, Amaury Dehecq, Emmanuel Trouve, Gilles Mauris, Noel Gourmelen, Flavien Vernier
2016 IEEE Geoscience and Remote Sensing Magazine  
5 Nowadays, data fusion constitutes the key subject in numerous applications of remotely sensed displacement 6 measurements, with the increasing availability of remote sensing data and the requirement of improvement 7 of the measurement accuracy. This paper addresses the current status and challenges in the fusion of re-8 motely sensed displacement measurements. An overview is given to discuss the remote sensing sources and 9 techniques extensively used for displacement measurement and the
more » ... t development and achievement of 10 displacement measurements fusion. Fusion between displacement measurements and integration of a geo-11 physical model are discussed. The fusion strategies and uncertainty propagation approaches are illustrated 12 in two main applications: 1) surface displacement measurements fusion to retrieve surface displacement 13 with reduced uncertainty in case of redundancy, with larger spatial extension or of higher level in case 14 of complementarity 2) surface displacement measurements fusion to estimate the geometrical parameters 15 of a physical deformation model in case of redundancy and complementarity. Finally, the current status 16 and challenges of remotely sensed displacement measurements fusion are highlighted. Moreover, some 17 potential ways are proposed to deal with heterogeneous data types and to assimilate remote sensing data 18 into physical models in order to realise near real time displacement monitoring. 19 1 Introduction 20 The surface of the Earth is deforming permanently due to mass transfer, either internal or external, either 21 natural or man-made activities. The displacement at the Earth's surface vary a lot in terms of spatial 22 extension, amplitude and temporal evolution. The investigation of the displacement at the Earth's surface 23 represents an essential part of geodesy and its quantification constitutes a major topic in the community 24 of geoscience, since it is of particular importance for natural hazards monitoring. For example, the dis-25 placement measurements and the deformation model inferred from these measurements provide crucial 26 information in order to avoid the installation of emergency shelter and reconstruction over affected areas 27 that will result in further damage in a later earthquake [1, 2]. Further, these sources of information enrich 28 the disaster early warning system in order to prevent the future natural hazards. Displacement measure-29 ments also present great potential for underground exploitation, bridges and dams sinking monitoring and 30 they are of particular interest in civil engineering [3, 4, 5, 6]. 31 At the end of the 20th century, the development of spatial geodetic techniques (optical & SAR imagery, 32 GPS) has allowed for drastic improvement of the spatial coverage, the resolution and the accuracy of 33 displacement measurements. Spaceborne optical and radar sensors observe the Earth's surface 34 continuously, across both space and time, but with limited flexibility in terms of revisit 35 time and acquisition geometry. Airborne optical and radar sensors provide displacement 36 measurements with limited spatial/temporal coverage, but improved flexibility in terms of 37 revisit time and acquisition geometry. Moreover, ground based optical and radar sensors, 38 with good flexibility in terms of revisit time and acquisition geometry, often give precise 39 information for small scale phenomena. Thanks to these techniques, spectacular results have been 40 obtained in numerous applications with displacement of various characteristics in terms of magnitude, 41 duration, spatial distribution, etc.: the study of subsidence in urban areas [3, 7, 8, 9, 10, 11], of the 42 co-seismic, inter-seismic and post-seismic motions [12, 13, 14, 15, 16], of glacier flows [17, 18, 19], of volcanic 43 deformation [20, 21, 22, 23], etc. Nowadays, the displacement maps obtained by remote sensing techniques 44 reach an accuracy within millimetres per year for deformation velocity and cover almost the whole land 45 of the Earth, including the non-instrumented remote areas and areas that do not have the necessary 46 financial means and human resources for ground instrumentation. They have also proven very useful for 47 regional studies. Moreover, due to the archiving system, a posteriori studies can be performed on areas 48 where an interesting phenomenon has been detected. We thus have access to the initial phase. Therefore, 49 remote sensing displacement measurements have obtained significant development in the past few years. 50 They are considered as the predominant source for the detection and the quantification of the terrestrial 51 deformation, from which geophysical models have been retrieved to further understand the deformation 52 source in depth and the physical process that induces the displacement observed from the Earth's surface. 53 To this end, a good knowledge of the reliability of the remotely sensed measurements, as well as 54 2 of the geophysical models accordingly obtained, is crucial for all the researches and applications that 55 use these sources of information. However, remote sensing displacement measurements are subject to 56 incompleteness and uncertainty. Uncertainty is also present in the geophysical model due to limited 57 knowledge about the phenomenon under observation and approximations made in the modelling, as well as 58 uncertainties associated with the displacement measurements used to constrain the model. A perspective 59 of significant reduction in the uncertainty of the displacement measurement appears thus with the 60 increasing availability of different types of remote sensing measurements and the blooming development 61 of displacement information extraction techniques. Thereby, the role of data fusion, making use of the 62 redundant and complementary displacement information brought by different sources, becomes more and 63 more important. Methodological development of the fusion of different types of displacement measurements 64 and of the integration of a physical model based on supercomputer facilities seems necessary to improve 65 the spatial extension and the accuracy of displacement measurements. In this context, this paper addresses 66 the current status and challenges of the fusion of remotely sensed displacement measurements. 67 This paper is organised as follows: In Section 2, remote sensing sources including optical, SAR images, in 68 situ GPS measurements and levelling sources, as well as displacement extraction techniques such as offset-69 tracking, differential interferometry (DInSAR) are introduced. Moreover, the uncertainty quantification 70 of measurements issued from these techniques is also discussed. In Section 3, the fusion of displacement 71 measurements and the integration of geophysical models are presented. The fusion issues are presented 72 through 2 main applications: from raw measurements to fused measurements and from measurements to 73 model parameters. Finally, in Section 4, the current status and challenges are highlighted and perspectives 74 to deal with heterogeneous data types and to assimilate remote sensing data into physical models are 75 proposed. 76 2 Displacement measurement data 77 Nowadays, SAR and optical images constitute the predominant remote sensing source for displacement 78 measurement, due to their high capacity in providing displacement measurement over large area and of 79 great accuracy. GPS and levelling measurements, thanks to their high precision, are also widely used as 80 complementary sources to remote sensing data. 81 3 2.1 Displacement extraction techniques 82 Two different families of technique have been developed to extract displacement information from SAR 83 or optical images: offset-tracking of amplitude SAR or optical images and DInSAR. Techniques in the 84 family of offset-tracking, based on the cross-correlation between the master image and the slave image, 85 provide two dimensional (2D) measurements (namely correlation measurements hereafter), with one hor-86 izontal component in the direction of the sensor motion and the other component in the perpendicular 87 direction in the horizontal plane for optical images and in the Line Of Sight (LOS) for SAR images. The 88 accuracy of these techniques is limited by the resolution of the images used, the stereoscopic effect and the 89 decorrelation. Numerous studies have confirmed that the displacement error is generally included between 90 tenth of pixel and one pixel [24, 25]. The best accuracy obtained is recorded as 1/30 pixel for SAR images 91 [26] and 1/200 of pixel for optical images [27] with careful data processing. The application of offset 92 tracking techniques is thus mainly determined by the resolution of the images used and the 93 magnitude of the displacement to measure. Therefore, they are commonly applied for large 94 displacement, e.g. glacier flow monitoring [28, 29, 30, 31] and strong earthquake measurement in the 95 field near the fault rupture [24, 12, 32, 33, 34, 35]. 96 DInSAR, on the other hand, makes use of the phase information included in a pair of SAR images 97 and allows for the measurement of the displacement occurred between the two acquisitions in the LOS 98 direction. Compared to offset-tracking, this technique requires strong coherence between two SAR 99 images, for which the geometrical and temporal baselines between the two acquisitions should 100 be as small as possible. Moreover, more complex processing steps such as the orbital, topographical 101 and atmospheric correction and phase unwrapping are necessary. In particular, phase unwrapping de-102 termining the success of the application of DInSAR, is difficult and delicate since the choice of the phase 103 unwrapping method depends on the nature of the interferograms to be processed. The problems mainly 104 encountered are the discontinuity of the coherent areas and the strong gradient of the displacement that 105 can cause potential aliasing problem. Today, no method seems fully operational. DInSAR has been widely 106 used to measure small displacements such as surface subsidence in urban area [7, 36, 37, 8], inter-seismic 107 deformation [14, 38, 16] or glacier flow [39, 40, 41, 18], with an average accuracy of centimetres. With the 108 increasing availability of SAR images, techniques such as Permanent Scatterer (PS) [42, 43, 44, 45] and 109 Small BAseline Subset (SBAS) [46, 47, 48, 49, 50] dealing with time series have been developed in order to 110 reduce the uncertainty of the displacement measurement and to get around of the principal limitations 111 of the conventional DInSAR technique. With these techniques and the availability of the X-band high 112 resolution images (TerraSAR-X, COSMO-SkyMed), precision on the order of millimetres per year 113 4 has been obtained for displacement rate. Recently, combination of these two techniques is performed in 114 order to further reduce the uncertainty of the displacement measurements and promising results have 115 been obtained [51, 52, 9]. Furthermore, multiple aperture InSAR (MAI) technique, based on 116 split-beam InSAR processing, has been developed in order to extract along-track displace-117 ment from DInSAR data [53, 54]. The along-track displacement obtained is consistent with 118 that obtained from offset-tracking. Note also that, in multitemporal InSAR processing, the 119 deformation velocity estimation can be strongly biased by the thermal dilation of the imaged 120 objects. Improvement of existing approaches and development of new approaches [55, 56, 57] 121 have been proposed to deal with this issue. With these approaches, it is possible to achieve 122 an extremely accurate monitoring of thermal dilation, up to a sensitivity on the order of 123 1 mm in the deformation measurement [56]. 124 Besides SAR and optical images, continuous GPS, as a complementary remote sensing source, is also 125 widely used in displacement measurement. Different from SAR/optical imagery, GPS provides the 3D 126 displacement (with 3 components: East, North, Up in the terrestrial reference) on a much sparse and 127 irregular spatial grid with temporal sampling every 5 minutes or even less. The uncertainty associated 128 with the GPS displacement measurement is sufficiently small, on the order of 5 -10 mm and 10 -20 mm 129 in horizontal and in vertical respectively [58]. Thanks to the dense temporal sampling, GPS allows us to 130 obtain time series for displacement varying over time, at the scale of days and years. GPS measurements 131 have been used in detection of tectonic activities like earthquake [59, 15]), volcano [60], glacier flow [61], 132 plate movement [62], etc. Moreover, levelling, the measurement of elevation difference between 2 points at 133 the Earth's surface, can also be considered as a precise method for vertical displacement measurement. It 134 has been used for displacement measurement for more than half a century [63, 64, 65, 15]). A precision on 135 the order of mm/yr has been reported for vertical displacement rate [66]. However, besides the punctuality 136 of the measurement, the major disadvantages of levelling also include the high cost and the large amount 137 of time needed for collecting the data over long distances or over a large network. 138 2.2 Uncertainty quantification 139 The sources of uncertainty in optical/SAR imagery are very complex: they come from different pertur-140 bations that take place along the electromagnetic wave propagation (e.g. atmosphere) and at the back-141 scattering surface (e.g. properties change during two acquisitions), as well as the noise generated in the 142 electronic processing. Moreover, imperfect displacement extraction technique (accuracy of the algorithm) 143 and pre/post-processing treatment (coregistration, geometrical correction, etc) also induce uncertainties
doi:10.1109/mgrs.2016.2516278 fatcat:bc2avngganasfc2t5v7xo6pzf4